JOBS WORKING PAPER Issue No. 5 Job Creation in the Private Sector Thomas Farole, Esteban Ferro, and Veronica Michel Gutierrez An Exploratory Assessment of Patterns and Determinants at the Macro, Sector, and Firm Levels © 2017 International Bank for Reconstruction and Development / The World Bank. 1818 H Street NW, Washington, DC 20433, USA. Telephone: 202-473-1000; Internet: www.worldbank.org. Some rights reserved This work is a product of the staff of The World Bank with external contributions. The findings, interpretations, and conclusions expressed in this work do not necessarily reflect the views of The World Bank, its Board of Executive Directors, or the governments they represent. The World Bank does not guarantee the accuracy of the data included in this work. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgment on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. Nothing herein shall constitute or be considered to be a limitation upon or waiver of the privileges and immunities of The World Bank, all of which are specifically reserved. Rights and Permissions This work is available under the Creative Commons Attribution 3.0 IGO license (CC BY 3.0 IGO) http://creativecommons.org/licenses/by/3.0/igo. Under the Creative Commons Attribution license, you are free to copy, distribute, transmit, and adapt this work, including for commercial purposes, under the following conditions: Attribution—Please cite the work as follows: Thomas Farole, Esteban Ferro, and Veronica Michel Gutierrez. 2017. “Job Creation in the Private Sector - An Exploratory Assessment of Patterns and Determinants at the Macro, Sector, and Firm Levels.” World Bank, Washington, DC. License: Creative Commons Attribution CC BY 3.0 IGO Translations—If you create a translation of this work, please add the following disclaimer along with the attribution: This translation was not created by The World Bank and should not be considered an official World Bank translation. The World Bank shall not be liable for any content or error in this translation. Adaptations—If you create an adaptation of this work, please add the following disclaimer along with the attribution: This is an adaptation of an original work by The World Bank. Views and opinions expressed in the adaptation are the sole responsibility of the author or authors of the adaptation and are not endorsed by The World Bank. Third-party content—The World Bank does not necessarily own each component of the content contained within the work. The World Bank therefore does not warrant that the use of any third-party-owned individual component or part contained in the work will not infringe on the rights of those third parties. The risk of claims resulting from such infringement rests solely with you. If you wish to re-use a component of the work, it is your responsibility to determine whether permission is needed for that re-use and to obtain permission from the copyright owner. Examples of components can include, but are not limited to, tables, figures, or images. All queries on rights and licenses should be addressed to World Bank Publications, The World Bank Group, 1818 H Street NW, Washington, DC 20433, USA; fax: 202-522-2625; e-mail: pubrights@worldbank.org. Project Number: P164622 Report Number: AUS22807 Images: © World Bank. Further permission required for reuse. ABSTRACT Why do some countries create more jobs than others? To consider this question, in this paper we focus on one of the most basic relationships, between growth and employment. In practice, the private sector responds very differently to growth (and decline) across countries. Understanding the patterns and drivers of private sector decisions to expand and shed jobs may be important to guide policy approaches for job creation. This paper analyzes the output-employment relationship in the context of business cycles at three levels: the macro-economy; industry (in manufacturing); and firms. The results highlight major differences in private sector job creation responsiveness to growth across stages of development and economic structures, but a critical finding is that economies (and firms) where formal sector job creation was more responsive to growth cycles generated more jobs overall. In addition, results from both the macro analysis and the sectoral analysis suggests significant complementarity between capital and labor. Finally, the findings may help to frame a broad policy agenda for job creation, including: macro-economic fundamentals, responsive labor markets, access to finance, competition, and a facilitative business regulatory environment. These are not surprising, but nevertheless frame a set of issues that could be explored in further research. 1 ACKNOWLEDGEMENTS This report was prepared by the World Bank Group’s (WBG) Jobs Group. The principal authors are Thomas Farole, Esteban Ferro, and Veronica Michel Gutierrez. We are grateful to Alvaro Gonzalez, Marek Hanusch, and Gonzalo Varela for their valuable comments and suggestions. We are also grateful to Alvaro Gonzalez, Hari Subhash, and Leonardo Lacovone for providing access to the dataset used in Section 3 of this paper. The publication of this report has been made possible through a grant from the World Bank’s Jobs Umbrella Trust Fund, which is supported by the Department for International Development/UK AID, and the Governments of Norway, Germany, Austria, the Austrian Development Agency, and the Swedish International Development Cooperation Agency. The report was prepared under project Firms, Access to Markets, Value Chains and Jobs (P164622). 2 CONTENTS ABSTRACT ............................................................................................................................................ 1 ACKNOWLEDGEMENTS ......................................................................................................................... 2 1. ECONOMIC GROWTH, THE PRIVATE SECTOR AND JOB CREATION – FRAMING THE ISSUE .................. 6 2. MACRO-LEVEL VIEW: THE LINK BETWEEN BUSINESS CYCLES AND EMPLOYMENT GROWTH .............. 9 Business cycles and employment growth – descriptive overview highlighting country variation ............... 9 Determinants of employment outcomes ................................................................................................... 18 3. INDUSTRY-LEVEL VIEW: ANALYSIS OF GROWTH AND EMPLOYMENT RELATIONSHIPS IN MANUFACTURING .............................................................................................................................. 21 ........................................................................................ 21 Relationship between growth and employment How are employment elasticities associated with overall job creation? ................................................... 26 4. FIRM-LEVEL VIEW: FIRM TYPES, REGULATION, AND JOB CREATION ................................................ 30 The regulatory environment and job turnover .......................................................................................... 31 5. CONCLUSIONS AND FUTURE RESEARCH .......................................................................................... 39 ANNEX A: OKUN’S LAW ...................................................................................................................... 42 BIBLIOGRAPHY ................................................................................................................................... 46 3 ABBREVIATIONS Most used abbreviations list: please keep relevant ones and add your own ADB Asian Development Bank ALMP active labor market program ASA advisory services and analytics CBO community-based organization CCSA cross-cutting solutions area CCT conditional cash transfer CEO chief executive officer CIS Commonwealth of Independent States CPF country partnership framework DAC Development Assistance Committee (OECD) DfID Department for International Development (United Kingdom) EBRD European Bank for Reconstruction and Development ECA Europe and Central Asia EU European Union FAO Food and Agriculture Organization of the United Nations FDI Foreign Direct Investment FCS fragile and conflict state FY fiscal year GDP Gross Domestic Product GIZ Deutsche Gesellschaft für Internationale Zusammenarbeit GNI Gross National Income GP global practice HIC high-income country HS household survey IAT informal apprenticeship training ICLS International Conference of Labour Statisticians ICR implementation completion report IDA International Development Association IE impact evaluation IDB Inter-American Development Bank IEG Independent Evaluation Group IGA income-generating activity ILC International Labour Conference IFC International Finance Corporation IILS International Institute for Labour Studies ILS International Labour Standards IMF International Monetary Fund IOE International Organization of Employers ITUC International Trade Union Confederation IZA Institute for the Study of Labor ICT information and communications technology ILO International Labor Organization 4 ISIC International Standard Industrial Classification KILM Key Indicators of Labor Market KM Knowledge Management LDC least developed country LFS labor force survey MDG Millennium Development Goal M&E monitoring and evaluation MFI micro-finance institution MoF Ministry of Finance MIS management information system MSE micro and small enterprise MSME micro, small and medium enterprise NIS National Institute of Statistics n.a. not applicable N/A, — not available NGO nongovernmental organization NEET not in education, employment, or training ODA official development assistance OECD Organization for Economic Co-operation and Development OOP out-of-pocket payments PER public expenditure review PEP public employment programme PES public employment services PPP purchasing power parity PHC population and housing census PRSP poverty reduction strategy paper RCT randomized control trial SME small and medium-sized enterprises SDG sustainable development goal SCD strategic country diagnostic SOE state owned enterprise STEP skills towards employment and productivity TFP Total Factor Productivity TOR terms of reference TVET Technical and Vocational Education and Training UN United Nations UNDP United Nations Development Programme UNIDO United Nations Industrial Development Organization USD United States Dollars VAT Value added Tax VET technical and vocational education and training VC value chain WBG World Bank Group WDI World Development Indicators WDR World Development Report WTO World Trade Organization 5 1. ECONOMIC GROWTH, THE PRIVATE SECTOR AND JOB CREATION – FRAMING THE ISSUE Listen to any policymaker’s speech or pick up any newspaper, and the urgency of job creation is clear. The World Development Report (WDR) on Jobs (World Bank, 2013) reports that more than 200 million people worldwide are unemployed, while another 2 billion working age adults remain outside the workforce. On top of this, the report estimates 600 million additional jobs will be needed over the next 15 years just to keep pace with new entrants to the labor market. Where will these jobs come from? While public sector employment has a role to play, the WDR argues that the private sector is the only sustainable engine of job creation in any economy, emphasizing the widely reported statistic that 9 out of every 10 jobs are in the private sector (World Bank, 2013). But this includes farmers, other self-employed, and informal workers, in addition to those that are wage workers in formal enterprises. Indeed, formal private sector jobs are the exception rather than the rule in many countries. Thus, the role of the private sector in job creation is likely to vary significantly depending on the structure of the economy and the nature of the private sector. It is also likely to vary significantly depending the policies governments pursue, which sets the conditions that facilitate or hinder private sector job creation. For this reason, understanding the patterns of employment across economies, time periods, and broad policy domains, is an important starting point to considering how to support robust private sector job creation. We can think of three broad (but not mutually exclusive) mechanisms that determine job creation in the private sector. First and foremost is growth – in response to increased aggregate demand new firms are established or when existing firms expand. Expanded output is usually a necessary condition, if not a sufficient one (output growth may be employment neutral if labor productivity increases at a corresponding rate or if firms substitute capital for labor). Second, for any given level of output, relative demand for labor may vary depending on the sectoral and enterprise structure of the economy (some sectors and enterprise types are more labor intensive than others. Finally, for any given level of output and sector/firm composition, the relative use of capital and labor (factor intensity) will determine the number of jobs created. This paper is intended as the first in a possible series of analyses on private sector job creation. We focus in this paper on the first of the mechanisms outlined above, analyzing one of the most basic relationships in economics, – between growth and employment – although we also explore briefly how compositional effects and factor intensity mediates the output-employment relationship. Subsequent papers in this series could probe some of the findings from this initial assessment as well as explore the other mechanisms outlined above. 6 This paper analyzes the output-employment relationship in the context of business cycles1, which allows for comparing across a large set of countries by measuring employment elasticities to growth and decline. A substantial literature exists job creation over business cycles. Most papers focus on which firm types drive job creation, including the long-standing debate on whether large or small and/or young or established firms create more jobs (Gertler and Gilchrist, 1994; Fort et al, 2013; Decker et al, 2014). Analysis taking the business cycles approach also tend to be single-country focused, most often using data from the United States or other high income countries, with relatively little analysis done on countries outside of the OECD (for an example of multi-country analysis outside the OECD, see Hanusch, 2012). Among the few large cross-country studies that follow a broadly similar approach, Freund and Rijkers (2014) study episodes of sharp unemployment reduction in 94 rich, middle-income, and transition countries. They find that while the business cycle is the biggest factor shaping unemployment trends, countries with a better regulatory environment better are less likely to have high unemployment and more likely to recover rapidly following a spike in unemployment. The main objective of this paper is modest – we aim to assess the relationship between growth and job creation across a broad cross-section of economies. The focus is on identifying and describing patterns in the relationship between growth and job creation over business cycles in different country contexts. But we also ask why is that some countries create more jobs than others during periods of growth. And so we begin to explore structural and policy-related factors that may explain the patterns of overall job creation and employment elasticity to growth. We can think of several factors that may be relevant in determining these patterns: structural and compositional effects; relative prices of capital and labor; market rigidities (business regulations, financial and labor markets); and uncertainties emanating from policy, institutions, and/or historical volatility. The hypothesis is that factors such as income levels, sectoral composition, capital intensity, and the structure of the enterprise sector and of labor markets will have significant impacts on how the private sector responds to create (shed) jobs in periods of growth (decline). However, controlling for structural factors, the hypothesis is that countries with more effective enabling environments (broadly defined) to support the private sector will create more jobs overall and exhibit employment outcomes that respond efficiently to business cycles (i.e. in terms of patterns of labor shedding and absorption). This hypothesis is explored through a combination of descriptive and regression analysis of cross-country data on employment outcomes at three levels: in the macro economy; at the meso (sector) level; and at the micro (firm) level. The paper is organized in four sections following this introduction. Section 2 takes a macro lens to the relationship between business cycles and employment outcomes, introducing the Okun coefficient as a key measure of the employment elasticity to growth. The Okun coefficient is derived from Okun’s Law, which refers to the empirical regularity found in the United States that seems to hold between the output and unemployment gaps, or between cyclical unemployment and cyclical output. We then test Okun’s Law across the global economy, where we find it holds well for high and upper middle-income countries. But we find that job creation is not associated with output changes in lower-middle income 1 We use the term ‘business cycles’ to refer to patterns of economic growth or decline; note that while the term ‘business cycle’ is often associated with quarterly periods, this paper we have data only in one-year intervals. 7 and low-income countries. This is perhaps not surprising, given that low-income countries are characterized by high levels of self-employment, and so labor markets may adjust to growth more through earnings than through job creation per se. Nevertheless, the findings are relevant as we also find that countries in which Okun’s law holds, have on average higher employment growth. Thus, understanding what determines Okun’s law may also help us discover what determines higher employment growth across countries. The third section moves to the sector level and carries out a more descriptive analysis with a cross- country dataset limited to the manufacturing sector. This helps to control for issues of sectoral composition and informality that introduce comparability problems in the macro analysis. We find that, controlling for these factors, job creation patterns in lower income countries respond equally well to output changes as in higher income countries (in fact, even more so). This corroborates the hypothesis that structural and compositional factors determine significantly the patterns observed in the macro data. The findings from the sectoral analysis also suggest that capital intensity is not necessarily a barrier to job creation. In this section, we also test to see if there is evidence of any systematic delays to the adjustment process at the sector of country-group level and find no evidence. Finally, we test for whether volatility in output impacts adjustment (as measured by contemporaneous elasticity of employment) and find supporting evidence at the sector level but not at the country level, suggesting that countries may follow an explicit or implicit ‘portfolio strategy’ in sectoral specialization that smooths volatility. Section 4 moves the lens to the firm level, exploiting firm microdata to understand how firms create jobs in response to output changes. We explore the links between the regulatory environment in which firms operate and job turnover by exploiting the observed industry-size variations through a difference- in-difference approach. Additionally, we explore firm characteristics that are associated with higher employment growth rates in the short run and the long run, and identify the firm types with the largest impact on employment growth. Understanding the determinants of the number of high growth firms across countries can help us understand that is needed to create the environment for these types of firms to flourish. We find mixed results, but overall macro, structural, and firm characteristics are most associated with job creation, while regulatory factors and institutions also mediate outcomes. Section 5 concludes and outlines possible paths for future research. 8 2. MACRO-LEVEL VIEW: THE LINK BETWEEN BUSINESS CYCLES AND EMPLOYMENT GROWTH BUSINESS CYCLES AND EMPLOYMENT GROWTH – DESCRIPTIVE OVERVIEW HIGHLIGHTING COUNTRY VARIATION The responsiveness of the labor market over business cycles has important implications for workers. Shocks to the economy cause output to fluctuate around the economy’s potential. These output movements cause firms to hire and fire workers, changing employment; changes in employment, in turn, move the unemployment rate in the opposite direction. But the nature and strength of this relationship varies. For example, in the US, the last three recessions (1990-91, 2001, and 2007-09) have been followed by "jobless recoveries". Jobless recoveries are periods following the end of recessions when output growth resumes but employment does not grow. In contrast, cyclical recoveries prior to 1990 were accompanied by prompt recoveries in employment and declines in unemployment. While labor market relationships with growth vary over time within countries, they vary even more, and more systematically, across countries. Figure 1 shows examples of the relation between GDP and employment growth for six different countries. In the top panel, we can see that in Sweden and the US there is very close relation between GDP growth and employment growth rates. It is also worth highlighting how these two economies are more stable than the others, particularly employment does not have major swings in either direction, with the exception of 2009 when the world financial crisis affected most developed countries. On the other hand, these two economies also show substantial differences in their response to the wild fluctuations in output that accompanied the 2009 crisis, with employment falling much less sharply in Sweden in 2009, but also rebounding much less sharply in 2010. This suggests that while the upper income status is part of the explanation for GDP and employment growth patterns, other factors also play a significant role, in this case potentially the level of safety nets available in Sweden versus the US, the relative share of private versus public employment, or the willingness of the private sector to maintain workers during a downturn. The middle panel shows two developing countries, Ecuador and Vietnam. It is clear that both GDP and labor markets are much more volatile than the countries in the top panel. Ecuador is an interesting example, which shows a low correlation between GDP and employment growth between 1995 and 2000; between 2000 and 2005 there is a high correlation, and between 2005 and 2010 there appears to be a negative correlation between the two indicators. In Vietnam, there also appears to be a negative correlation between GDP and employment growth. The final panel shows the relation between GDP and employment growth for Burundi and China. In these countries, there appears as if there is no correlation between GDP and employment growth. In the case of Burundi this is partially explained by the fact that nearly 90 percent of the labor force is self-employed or otherwise informal. In this context, employment responses to growth (or decline) are unlikely to register as significant (this issue will be discussed in more detail later in this section). 9 Figure 1 Relation between GDP and Employment Growth GDP and Employment Growth (detrended) GDP and Employment Growth (detrended) Sweden United States 2 5 0 Growth Rate (%) Growth Rate (%) 0 -2 -5 -4 -10 -6 1995 2000 2005 2010 2015 1995 2000 2005 2010 2015 Year Year GDP Growth Employment Growth GDP Growth Employment Growth GDP and Employment Growth (detrended) Vietnam 2 1 Growth Rate (%) 0 -1 -2 1995 2000 2005 2010 2015 Year GDP Growth Employment Growth GDP and Employment Growth (detrended) GDP and Employment Growth (detrended) Burundi China 10 4 5 Growth Rate (%) Growth Rate (%) 2 0 0 -5 -10 -2 1995 2000 2005 2010 2015 1995 2000 2005 2010 2015 Year Year GDP Growth Employment Growth GDP Growth Employment Growth Okun’s Law Typically, growth slowdowns coincide with rising unemployment, and vice versa. This negative correlation between GDP growth and unemployment has been named “Okun’s law,” after the economist Arthur Okun who first documented it in the early 1960s. Part of the enduring appeal of 10 Okun’s law is its simplicity, since it involves two important macroeconomic variables. Additionally, the relationship appears to enjoy empirical support. Ever since Okun (1962) tested it empirically, the relationship has been examined by a number of economists including, inter alia, Smith (1975), Gordon (1984), Knoester (1986), Kaufman (1988), Prachowny (1993), and Weber (1995). While most these studies examined its validity for the United States economy, the relationship has also been tested for other countries, e.g., Knoester (1986) Kaufman (1988), and Moosa (1997). Also, Ball, Leigh and Loungani (2016), show that Okun’s Law has held up well for a set of 20 advanced economies. Although the results generally support the empirical validity of the relationship in the sense of finding a significantly negative coefficient on cyclical output, the so-called Okun coefficient; nevertheless, the authors find significant cross-country differences in the magnitude of the coefficients, with larger coefficients found for the United States and Canada than for Europe and Japan. The authors attribute the difference in the elasticity across countries to differences in the rigidity of the labor markets. We expect to that Okun’s law will be applicable in developing countries; however, we can expect to find that the Okun coefficient will be significantly different across countries. Since labor markets differ in regulations, structure, and each country has a very different economic structure in terms of what they produce and in the income level of their populations we also expect that there are significant differences in the way labor markets react to changes in business cycles. For example, in many low- income countries the large majority of the working age population is self-employed, either in subsistence agriculture or some informal non-farm activity. While these are accounted for as ‘jobs’ in the employment data, the reality is that they will not come and go in periods of economic growth and decline; rather, the adjustment to business cycles is likely to be observed through earnings, working hours, and/or shifts in employment within the informal sector. In this section, we examine—among other employment outcomes—how the Okun coefficient behaves for developing and developed countries. Since Okun’s Law has not been studied extensively in the context of developing countries (although see Hanusch, 2012, for one example), and is a less well known concept, much of this subsection explains how the Okun coefficient is estimated and how it differs between countries of different income levels. We then explain why some countries’ labor force are more responsive to changes in outcome than others. Since we are particularly interested in how private sector contributes to job creation, we explore employment in the formal sector as well as private sector determinants of the Okun coefficient. The motivation is two-fold. First, developing countries account for a large, and growing, share of the global labor force. Hence, understanding the determinants of labor market outcomes in these countries is important. There is ample evidence that job creation contributes to individual and social welfare, whereas unemployment and job loss are associated with persistent loss of income, health problems, and breakdown of family and social cohesion (see the World Bank’s World Development Report on “Jobs” (2013)). A second motivation is to probe the common perception that labor market outcomes in developing countries reflect mostly structural factors rather than short-run cyclical fluctuations. Whether this perception is correct has important policy implications. If cyclical fluctuations account for a substantial part of labor market developments, macroeconomic stabilization policies—such as central bank actions, countercyclical fiscal policies and prudential policies to mitigate financial crises—gain in importance relative to structural policies (e.g. improving education and skills of the labor force). 11 There are different versions on how to estimate Okun’s coefficient. However, Okun’s two relationships, arise from the observation that more labor is typically required to produce more goods and services within an economy. More labor can come through a variety of forms, such as having employees work longer hours or hiring more workers. To simplify the analysis, Okun assumed that the unemployment rate can serve as a useful summary of the amount of labor being used in the economy. A high rate of unemployment, Okun reasoned, would typically be associated with idle resources. In such a circumstance, one would expect the actual rate of output to be below its potential. A very low rate of unemployment would be associated with the reverse scenario. The original version of Okun’s law compares the gap between actual and potential (or natural) rate of unemployment and the gap between actual output and potential output. Alternatively, we estimate the difference version of Okun’s law which has the advantage that there is no need to determine the natural output and employment which a non-trivial exercise. This version of Okun’s Law posits a relationship between the changes in the unemployment rate and the growth rate of output and is defined as follows: ∆"#,% = '∆(#,% + *#,% (1) where ∆"#,% is the change in the unemployment rate in country i in year t. ∆(#,% is the change in output in country i in year t, and *#,% is the error term. Equation (1) may be modified by including a constant term and introducing some limited dynamics to obtain ∆"#,% = + + ,∆"#,%-. + '∆(#,% + /#,% (2) The addition of lagged change in unemployment is required to remove serial correlation which often arises if the static equation (3) is used.2 This model is estimated using a rolling OLS with a window of 10 years. A rolling regression estimates a specific relationship over many different sample periods. Each regression produces a set of estimated coefficients. If the relationship is stable over time, then the estimated coefficients should be relatively similar from one regression to the next. Variations in the relationship will appear as sizable movements in the estimated coefficients. Each rolling regression is estimated based on 10 year of data points. Thus, the first rolling regression would estimate the values of + and ' from equation (2), using the sample period for the first 10 years of available data for country i. for example from 1991 to 2000. The sample period is then moved forward one year, and the regression is re-estimated to produce a second set of estimates of + and ' , using data 2 It must be emphasized, however, that following Hendry and Mizon (1978) and, more recently, Mizon (1995) serial correlation is viewed as indicating misspecification, and so the model is respecified instead of adopting the faulty alternative of correcting for serial correlation. In a paper with the provocative title ‘‘A simple message for autocorrelation correctors: Don’t,’’ Mizon (1995) asserts that correction for autocorrelation may yield inconsistent estimates, that it is generally invalid, and that it cannot be justified on the grounds of ‘‘robustifying’’ estimation against the presence of residual serial correlation. In a study of Okun’s law, Weber (1995, p. 440) also found the static regression to exhibit first-order serial correlation which he considered to be ‘‘evidence for misspecified dynamics,’’ and so he introduced dynamic equations. In our case, the limited dynamics turns out to be adequate because of the use of annual data. 12 from 1992 to 2001. This process is repeated until the final estimates are made using the entire sample period available for country i. As a robustness check on our results we also estimate the original version of Okun’s law using different methods to estimate the natural rate of unemployment and output. More details on these estimations can be found in the Annex. Applying Okun’s Law to The Rest of The World The idea is to explore how Okun’s law behaves in different countries and to learn how this relationship can help us understand private sector job creation around the world. Countries with higher absolute Okun coefficients will be able to absorb more workers when the economy is expanding, whereas in economic downturns, firms in countries with high absolute Okun coefficient are able to lay off workers to adjust to the new macroeconomic conditions in the market. On the other hand, an Okun coefficient closer to zero would imply that the labor market is unresponsive to economic expansions or contractions. We estimate the Okun coefficient using unemployment rate and GDP data from the World Bank’s World Development Indicators. The sample consists of 176 countries from different income groups and regions for the period 1993 until 2015. To motivate this research, Figure 2 shows, the relation between changes in the unemployment rate and real output growth for several countries, which belong to different income groups, regions, and economic structure. It is evident that Okun’s law is not stable across countries. Even though Okun’s law holds—for the most part—in the United States across time, Okun’s coefficient is not constant across countries. Figure 2 shows that the Sweden’s and United States’ (two high income countries) labor markets are more responsive to changes in real GDP than Ecuador’s and Thailand’s labor markets (two upper middle income countries). Honduras (lower middle-income country) and Haiti (low-income country) have labor markets that are barely responsive to GDP growth nevertheless, the slopes of the regression lines are negative. On the other hand, labor markets in Cameroon (lower middle-income country) and Rwanda (low-income country) are non-responsive to GDP growth, the regression line even has a slightly positive slope. Furthermore, in the US and Thailand for example, the observations are well represented by the fitted line whereas for the other countries the observations are more distant to the fitted line. 13 Figure 2 Okun’s law: examples of country variation Okun's Law - GDP Growth and Unemployment Okun's Law - GDP Growth and Unemployment Sweeden United States 4 4 Change in Unemployment Rate Change in Unemployment Rate 2 2 0 0 -2 -2 -5 0 5 10 -5 0 5 10 GDP Growth GDP Growth Okun's Law - GDP Growth and Unemployment Okun's Law - GDP Growth and Unemployment Thailand Ecuador 4 4 Change in Unemployment Rate Change in Unemployment Rate 2 2 0 0 -2 -2 -10 -5 0 5 10 -5 0 5 10 GDP Growth GDP Growth Okun's Law - GDP Growth and Unemployment Okun's Law - GDP Growth and Unemployment Honduras Cameroon 4 4 Change in Unemployment Rate Change in Unemployment Rate 2 2 0 0 -2 -2 -5 0 5 10 -5 0 5 10 GDP Growth GDP Growth 14 Okun's Law - GDP Growth and Unemployment Okun's Law - GDP Growth and Unemployment Rwanda Haiti 4 Change in Unemployment Rate Change in Unemployment Rate 4 2 2 0 0 -2 -2 -5 0 5 10 -5 0 5 10 GDP Growth GDP Growth In these examples, a pattern develops where countries with higher levels of development tend to have more responsive labor markets to changes in GDP growth compared to those in less developed countries. In the following section, we explore the relation between changes in the unemployment rate and GDP growth by country groups including income and economic structure. Figure 3 shows how Okun’s law performs in countries with different income levels. We can clearly see from the graphs that the Okun coefficient decreases (absolute terms) with the level of income. Unemployment rates in high- income countries improve on average by 0.21 for every percentage point increase in GDP.3 The responsiveness of the unemployment rates to changes in GDP decreases by income groups. The average Okun coefficient for upper-middle income countries is -0.08, -0.03 for lower middle-income countries, and -0.005 for lower income countries. The statistical significance of the coefficients and the R2 for the regressions also decreases as the level of income decrease.4 As suggested previously, these findings on income groupings are not at all unexpected. In most low- income countries, very few are unemployed – lack of sufficient safety nets means they cannot afford to be. But rather a large share of workers are self-employed, either in subsistence or basic agriculture, or in non-farm self-employment. In these environments, adjustment to upturns and downturns in the economy are likely to take place through earnings rather than employment per se. On the other hand, Figure 3 highlights significant dispersion around the fitted lines. This means that even among countries within similar income groupings, there exist large differences in the responsiveness of job creation to growth. For example, the Okun coefficient is 0.85 in Poland, but just 0.03 in Hungary; it is 0.43 in Bolivia and 0.38 in Chile but just 0.03 in Peru; similarly, while the Okun coefficient is nearly zero in Vietnam it is 0.32 in Indonesia. 3 The Okun’s coefficient for the United States is 0.44 but with a constant term equal to 1.2. This is higher than the one reported by Knoteck (2007); however, the sample period is different. 4 Annex Figure 1 shows the distribution of Okun’s coefficients by income groups. 15 Figure 3 Okun’s law by Income Group Okun's Law - GDP Growth and Unemployment Okun's Law - GDP Growth and Unemployment High Income Countries Upper-Middle Income Countries 4 4 Change in Unemployment Rate Change in Unemployment Rate 2 2 0 0 -2 -2 -4 -4 -10 0 10 20 -20 -10 0 10 20 GDP Growth GDP Growth Okun's Law - GDP Growth and Unemployment Okun's Law - GDP Growth and Unemployment Lower-Middle Income Countries Low Income Countries 4 4 Change in Unemployment Rate Change in Unemployment Rate 2 2 0 0 -2 -2 -4 -20 -10 0 10 20 -20 -10 0 10 20 GDP Growth GDP Growth Okun’s law also varies greatly by the economic structure of each country (Figure 4). The less sophisticated an economy is the less responsive is the labor force to changes in GDP. The negative relationship between unemployment rates and GDP growth is stronger in industrial, service intensive, tourism intensive, and high tech product intensive economies. In contrast, economies, which depend on natural resources and agriculture, have a regression line with a slope that is nearly zero.5 Again, some of this is likely related to the structure of employment (reliance on self-employment), at least in the case of agriculturally-dominated economies, or more broadly in limited share of formal employment in these countries. 5 To define the economic structure of a country we use the percentage of natural resources, agriculture, industry, services as a percentage of GDP and tourism and high-tech exports as percentage of total exports. For each of these measures we compute the 80th percentile per year and estimate the difference between each country’s measure and the 80th percentile. We then rank these differences; the largest value determines the group to which the country belongs in. 16 Figure 4 Okun’s law by Economic Structure Okun's Law - GDP Growth and Unemployment Okun's Law - GDP Growth and Unemployment Natural Resource Intensive Countries Agricultural Countries 4 4 Change in Unemployment Rate Change in Unemployment Rate 2 2 0 0 -2 -2 -4 -4 -20 -10 0 10 20 -20 -10 0 10 20 GDP Growth GDP Growth Okun's Law - GDP Growth and Unemployment Okun's Law - GDP Growth and Unemployment Industrial Countries Service Intensive Countries 4 4 Change in Unemployment Rate Change in Unemployment Rate 2 2 0 0 -2 -2 -4 -4 -20 -10 0 10 20 -10 -5 0 5 10 15 GDP Growth GDP Growth Okun's Law - GDP Growth and Unemployment Okun's Law - GDP Growth and Unemployment Tourism Intensive Countries High Tech Countries 4 4 Change in Unemployment Rate Change in Unemployment Rate 2 2 0 0 -2 -2 -4 -4 -10 0 10 20 -10 -5 0 5 10 15 GDP Growth GDP Growth Relationship Between Okun’s Coefficient and Aggregate Job Creation That there are differences in the responsiveness of private sector job creation to economic growth only matters if this relationship is associated with greater overall job creation. In this section, we explore the correlation between employment growth and the Okun coefficients estimated for each country. We expect to find that a larger negative Okun’s coefficient is associated with higher employment growth 17 since the labor force moves more freely with GDP movements. We confirm our hypothesis with a set of regressions where the dependent variables are employment growth rate in the formal sector as well as the number of formal jobs added from year to year. We control for country and time fixed effects. Results are displayed in Table 1. We use the three different estimates for unemployment Okun as described in the previous section and in the Annex. Okun’s coefficient does appear to be associated with higher levels of formal employment growth. The results highlight the importance of this relation and supports the idea that the Okun coefficient should be analyzed as a relevant employment outcome. Table 1 Relation between Okun’s Coefficient and Employment VARIABLES formal employment growth formal jobs added Estimation Method OLS GMM OLS GMM OLS GMM OLS GMM OLS GMM OLS GMM 1 2 3 4 5 6 7 8 9 10 11 12 okun (difference) -0.041 -0.093 -0.616 -0.950 [0.012]*** [0.029]*** [0.192]*** [0.407]** okun (gap - detrend) -0.026 -0.057 -0.407 -0.858 [0.015]* [0.027]** [0.210]* [0.305]*** okun (gap - moving average) -0.020 -0.033 -0.421 -0.687 [0.017] [0.029] [0.178]** [0.246]*** formal employment growth (t-1) -0.385 -0.397 -0.397 [0.038]*** [0.037]*** [0.040]*** formal jobs added (t-1) -0.269 -0.268 -0.247 [0.044]*** [0.046]*** [0.043]*** Constant 0.020 0.046 0.023 0.037 0.042 0.058 10.844 13.850 10.899 13.422 11.027 13.611 [0.006]*** [0.043] [0.007]*** [0.016]** [0.039] [0.045] [0.122]*** [0.577]*** [0.123]*** [0.587]*** [0.136]*** [0.506]*** Observations 1,711 1,549 1,555 1,400 1,593 1,433 1,272 814 1,154 729 1,182 750 R-squared 0.148 0.144 0.143 0.820 0.822 0.824 Number of id 136 136 136 127 125 126 Robust standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1 DETERMINANTS OF EMPLOYMENT OUTCOMES Thus far we have documented important differences in Okun’s coefficients across income groups and across different economic structures, and have shown that these differences matter from the perspective of overall job creation. What determines the responsiveness of employment to changes in GDP? Is there a common pattern across countries that can help us explain employment outcomes across countries? We identify possible determinants of employment outcomes, including indicators of the labor market structure, macroeconomic indicators that measure economic stability and investment, indicators of labor regulation, indicators of quality of institutions. These indicators come from the World Development Indicators (WDI) database. We also include several indicators that measure the ease of doing business in the country. These indicators are based on the World Bank Group’s Doing Business report. For these measures to be comparable across countries and time, they are recomputed as the distance to the frontier. A measure of 100, is the best or the frontier, and a measure of 0 is the furthest from the frontier.6 We expect that macroeconomic variables such as GDP per capita, GDP growth, gross fixed capital formation, foreign direct investment, openness, and domestic credit to the private sector should all have a positive association with employment growth (negative relation with the Okun coefficient). A better business environment (i.e., time to start a business and the global measure of doing business) 6 For this section on the determinants of employment outcomes, we omit from the sample low income countries because from the previous section we learned there is no statistical relation between changes in unemployment rates and the rates of growth of GDP. Including these countries in the sample only adds noise to our regressions without providing any additional benefit. 18 and institutions (i.e., rule of law) should have a positive effect on employment growth and a negative effect on the Okun coefficient. Among the measures of labor market rigidity, we expect a more flexible probation time for new hires will have a positive effect of employment growth (negative on the Okun coefficient) whereas a higher severance pay for redundancy dismissal will negatively affect employment growth and positively affect the Okun coefficient. Finally, as mentioned earlier, the bigger the informal sector of a country, measured by the percentage of the labor force that is self-employed, the slower we expect employment to grow (negative coefficient on employment growth) as well as a slower reaction of the labor force to react to changes in business cycles (positive coefficient on the Okun regression). Table 2 shows the results of regressions where the dependent variables are the employment outcomes of interest, including employment growth rate, the change in the number of jobs in the formal sector, and the Okun coefficients found for each country. The independent variables are described above. We make use of two specifications. The first includes country and year fixed effects, whereas the second specification only includes year effects. Instead of country fixed effects, we use additional controls that only vary across countries but do no change over time, such as the initial labor force and the initial level of development measured by a country’s GDP per capita. We include this second specification because we want to exploit the variation across countries and much of this variation is absorbed by the fixed effects used in the first specification. The most robust results suggest that better employment outcomes are associated with higher rates of GDP growth, higher levels of economic development, and lower levels of labor informality. The Okun coefficient is negative and statistically significant in the first set of regressions, indicating that a more responsive labor force results in higher employment growth, and a higher number of created jobs. Among other country characteristics that are significantly associated with higher job creation are higher FDI inflows and, counterintuitively, a more closed economy as measured by its openness (the sum of a country’s exports and imports as percentage of GDP) (see columns 2 and 5). Better institutions, measured by the rule of law of a country, is associated with a more responsive labor force to changes in business cycles (columns 3 and 6). Surprisingly our results show that the ease to hire and fire workers does not appear to influence the employment outcomes. Our proxies for the rigidity of the labor market, the maximum time for a probation period during the hiring process and the severance pay for redundancy dismissal for 1 year of service, are not statistically significant except for severance pay variable in column 6, which has the opposite sign than expected. The results suggest that higher severance pay is associated with a more responsive labor force, even though a higher severance pay means that it is more burdensome and expensive for firms to fire workers. Previous literature suggest that the responsiveness of labor markets could depend on regulations governing labor and product markets. For instance, in discussing hiring and firing regulations in Middle Eastern and North African countries, Ahmed, Guillaume, and Furceri (2012) argue that such regulations can discourage “firms from expanding employment in response to favorable changes in the economic climate.” That is, greater employment protection can dampen hiring and firing as output fluctuates, reducing the employment responsiveness. We find little association between the Okun coefficient and aggregate measures of labor market flexibility. Our results also suggest that private sector investment and private sector access to credit do not influence employment outcomes. In this section, we learned that Okun’s law is applicable to other countries; however, the Okun coefficient varies greatly across countries. The level of income and the economic structure of a country 19 determine how responsive the labor force is to changes along the business cycles. Because the Okun coefficient is correlated to higher employment growth, it is important to analyze it as an employment outcome as well as a determinant of employment growth. Unfortunately, we find that neither job creation nor the elasticity of unemployment to GDP is determined by private sector characteristics such as private sector investment and private sector access to credit but, mainly by macroeconomic and structural determinants such as GDP growth and the level of labor informality in a country. Table 2 Determinants of Employment Outcomes 1 2 3 4 5 6 employment change in okun employment change in okun VARIABLES growth rate formal jobs (difference) growth rate formal jobs (difference) okun (difference) -0.063 -0.915 -0.025 0.050 [0.038]* [0.403]** [0.022] [0.298] polulation growth rate -0.040 0.355 -0.016 0.008 0.368 0.017 [0.020]* [0.336] [0.033] [0.006] [0.072]*** [0.023] GDP growt rate 0.005 0.069 0.001 0.005 0.055 0.000 [0.001]*** [0.016]*** [0.002] [0.001]*** [0.014]*** [0.004] log GDPpc -0.119 -2.014 -0.203 -0.022 0.308 0.039 [0.052]** [0.799]** [0.092]** [0.020] [0.204] [0.049] informality rate -0.011 -0.090 -0.003 -0.018 -0.032 0.003 [0.003]*** [0.034]*** [0.003] [0.005]*** [0.006]*** [0.002]* domestic credit to private sector (% of GDP) 0.000 0.008 0.001 0.000 -0.011 -0.000 [0.000] [0.008] [0.001] [0.000] [0.005] [0.001] gross fixed capital formation, private sector (% of GDP) 0.002 -0.010 0.001 0.000 -0.010 0.002 [0.001] [0.016] [0.002] [0.001] [0.014] [0.002] foreign direct investment, net inflows (% of GDP) 0.000 0.027 0.001 0.001 0.019 -0.001 [0.001] [0.016]* [0.002] [0.001] [0.015] [0.002] openess -0.001 -0.019 -0.001 -0.000 -0.002 0.000 [0.000] [0.006]*** [0.001] [0.000] [0.002] [0.000] time to start a business (distance to frontier) 0.000 0.006 0.000 0.000 0.000 0.000 [0.000] [0.004] [0.001] [0.000] [0.003] [0.001] rule of law 0.021 -0.035 -0.102 -0.004 -0.306 -0.162 [0.031] [0.538] [0.049]** [0.023] [0.235] [0.062]*** average Doing Buiness global (distance to frontier) 0.001 0.010 0.004 [0.002] [0.016] [0.003] log max length of probation period (months) -0.009 -0.257 0.035 [0.010] [0.123]** [0.024] Severance pay for redundancy dismissal (for a worker -0.002 -0.034 -0.010 with 1 year of tenure, in salary weeks) [0.002] [0.019]* [0.004]** log (employment at t0) 0.141 1.997 0.139 [0.138] [1.513] [0.307] log (GDP at t0) 0.002 -0.291 0.105 [0.025] [0.178] [0.085] log (GDPpc at t0) -0.020 -0.289 -0.131 [0.033] [0.274] [0.114] log (labor force at t0) -0.135 -0.805 -0.236 [0.137] [1.541] [0.310] Constant 1.464 32.210 1.672 0.193 0.902 -0.962 [0.460]*** [7.368]*** [0.843]** [0.143] [1.614] [0.409]** Fixed effects c, y c, y c, y year year year Observations 405 308 413 309 231 317 R-squared 0.417 0.822 0.660 0.145 0.779 0.163 Robust standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1 20 3. INDUSTRY-LEVEL VIEW: ANALYSIS OF GROWTH AND EMPLOYMENT RELATIONSHIPS IN MANUFACTURING Given the significant differences in economic structures across countries, some of the key findings from the previous section may be explained by the fact that the employment responses to growth in economies dominated by agriculture and informality cannot be compared with those where most workers are wage employed. Therefore, in this section we provide some initial descriptive analysis carried out at the sectoral level, within manufacturing. This helps to control both for sectoral variation and mitigates issues related to informality. On the other hand, by restricting the analysis to within the sector, we are unable to capture spillover effects (e.g. multipliers arising from supply chain linkages and other indirect effects). The dataset used for the analysis is drawn from the Industrial Statistics Database of the United Nations Industrial Development Organization (UNIDO). It provides industrial statistics reflecting characteristics of the units engaged in a class of industrial activities, which are arranged according to the International Standard Industrial Classification of All Economic Activities (ISIC) summarized to 2 digits. Data is reported by country, year and ISIC code. It includes 170 countries and 52 years (1963-2014), although coverage varies by country. Specifically, this analysis comprises an unbalanced panel that uses 23 classes of industrial activity within the manufacturing sector, defined in terms of the ISIC Revision 3. It makes use of three statistics: number of employees, value added, and gross fixed capital formation, and draws on previous analysis from Gonzalez, Iacovone, and Subhash (2013). RELATIONSHIP BETWEEN GROWTH AND EMPLOYMENT As a first step, we assess correlations between employment growth and growth in output and value added. As shown in Table 3, moderate correlation is evident between employment growth and output growth in the manufacturing sector. Not surprisingly, this correlation increases when using value added rather than simple output as the measure for sectoral growth. We also assess whether the employment relationship with output and value added is stronger in a contemporaneous measure (same year) or if employment outcomes are lagged by one or two years. As Table 2 shows, the relationship diminishes to insignificance as soon as we start to lag the impact, suggesting the contemporaneous effect is more appropriate for considering sectoral responses to output and value added growth. For the remaining analysis in this section we, therefore, focus only on the contemporaneous relationship of employment growth to value added growth. Table 3 Correlation coefficient of manufacturing sector employment with output and value added Contemporaneous One-year lag Two-year lag Output 0.230 0.0034 0.0035 Value Added 0.357 -0.0037 0.0051 21 How do these relationships vary by income groups? Table 4 shows interesting results. Most striking is the extremely high correlation between job growth and value added growth in low income countries. This result is striking not only because the correlation is so high (this might, in fact, be expected in economies which are likely to be relatively low wage and labor intensive) but because the relationship is so dramatically different from what we saw with the macro look in Section 2, where low-income countries exhibited virtually no correlation between GDP growth and employment growth. The finding here suggests that the macro result is indeed shaped largely by the dominance of self-employment, particularly in agriculture, in these economies. Whereas once we focus on a sector that is largely formal, the tight relationship between growth and job creation becomes clear. The other interesting finding in Table 4 is the fact that high income countries also show a relatively strong correlation between value added growth and job creation. One reason for this may be that although high income countries have relatively limited direct employment creation through value added growth, they have much higher indirect job creation through supply chain spillovers, while low and middle income countries may have less integrated domestic supply chains, limiting potential employment multipliers through growth. While employment-output relationships are relatively strong in high income countries, they are, by contrast, much weaker in lower middle income and (especially) upper middle income countries. While we have no data to indicate why this may be the case, it may be that the middle-income category contains a diverse set of economic structures and contexts. It may also have something to do with the nature of the industrial transformation (see results below on how similar patterns emerge when we go from low to mid-level to high levels of capital intensity). Table 4 Correlation coefficient of manufacturing sector employment and value added by income level Income level Value added (contemporaneous) Correlation coefficient N Low 0.86685 4238 Lower middle 0.23576 12,833 Upper-middle 0.17430 16,214 High 0.46981 36,020 Figure 5 mirrors the descriptive analysis from Figure 1, illustrating annual trends in value added and employment growth (detrended) – in this case with data from the manufacturing sector. The analysis compares many of the same countries as in Figure 1, although manufacturing sector data is not available for all countries presented previously. While the relationship between value added and employment growth in Sweden’s manufacturing sector matches almost exactly with the overall economy picture shown in Figure 1, the patterns in the US are remarkably different (both from Sweden and from the US overall economy). Here we see job creation trending with value added growth in the two tails of the period, but diverging sharply (and inexplicably) from the mid-1990s through the mid-2000s. As might be expected from natural resources-based economies, both Ecuador and Cameroon show, for most of the period, relatively modest shifts in employment growth despite much sharper swings in value added growth. Finally, we look at manufacturing-oriented economies in latter stages of structural transformation. In the case of Turkey, job creation appears to respond fluidly to changes in value added, with greater responsiveness on the upside than on the downside since 2000. For most of the period (at least since 1995), China similarly shows strong elasticity of employment to manufacturing value added, 22 which stands in contrast the much weaker relationship between jobs and growth in the overall economy (Figure 1). Figure 5 Manufacturing value added and employment growth – country examples Value Added and Employment Growth (detrended MVA) Value Added and Employment Growth (detrended MVA) Sweden USA 5 20 Growth Rate (%) Growth Rate (%) 0 0 -20 -40 -5 1995 2000 2005 2010 2015 1990 1995 2000 2005 2010 Year Year Value Added Growth Employment Growth Value Added Growth Employment Growth Value Added and Employment Growth (detrended MVA) Value Added and Employment Growth (detrended MVA) Ecuador Cameroon 100 40 50 20 Growth Rate (%) Growth Rate (%) 0 0 -50 -20 -40 -100 1990 1995 2000 2005 2010 1980 1985 1990 1995 2000 Year Year Value Added Growth Employment Growth Value Added Growth Employment Growth Value Added and Employment Growth (detrended MVA) Value Added and Employment Growth (detrended MVA) Turkey China 40 40 20 20 Growth Rate (%) Growth Rate (%) 0 0 -20 -20 -40 -40 1995 2000 2005 2010 2015 1990 1995 2000 2005 Year Year Value Added Growth Employment Growth Value Added Growth Employment Growth In Figure 6, we illustrate differences in the growth-employment relationship in manufacturing subsectors; the table is organized according to the standard categorization of ISIC 2-digit sectors. We see 23 very large variations across sectors, which might be expected. Sectors with the highest correlations are: furniture, wood products, and machinery and equipment, which tend to be relatively labor intensive, although also somewhat capital intensive. By contrast, sectors where job creation is least responsive to growth patterns include one of the most labor intensive sectors (wearing apparel) along with sectors that tend to be capital intensive, including: radio, television and communication equipment; petroleum, and chemicals. Figure 6 Correlation coefficient of manufacturing sector employment and value added by income level One factor that may explain country differences, as well as subsectoral differences is the relative capital intensity of manufacturing activities. We would assume that more capital-intensive production would be associated with a lower correlation between value added growth and jobs growth. Table 5 shows the results of correlations using two different measures of capital intensity (capital to labor ratio and capital to value added ratio). The results confirm the expected negative relationship between capital intensity and employment responsiveness to growth, with broadly similar trends using both measures. However, it the decline in correlation between low and high levels of capital intensity is relatively minimal. Table 5 Correlation coefficient of manufacturing employment and value added by capital intensity Capital to labor ratio Capital to value added ratio corr_coeff _N corr_coeff _N Low 0.457431 25107 Low 0.577714 26461 Middle 0.354886 28447 Middle 0.466883 25972 High 0.35204 16585 High 0.357352 17706 We also check again whether it may be the case that some country groups or sector types (capital intensity of sectors) may adjust differently if, for example, structural factors delay but do not prevent 24 the adjustment process. Figure 7 presents the correlation between value added growth in time t and employment growth in different stages: contemporaneous (time t) as well as lagged by 1-3 years. The basic correlation analysis does not give any hint of systematic adjustment delays. In fact, all country grouping and capital intensity grouping react in broadly the same way, with high contemporaneous correlation, rapid degradation of the correlation (and turning negative) time t+1, low but positive in time t+2, and then further fading of the relationship. Figure 7 Correlation coefficient of manufacturing sector employment growth in time t, t+1, t+2, and time t+3 to value added growth in time t Finally, we ask whether volatility in output may affect adjustment. To the degree there are restrictions to the flexibility of hiring and firing or even just transaction costs involved in the process (which will always be the case), we can imagine firms will be more hesitant to hire in situations where output tends to vary significantly year-on-year. By contrast, in countries and sectors where demand and output are relatively stable, firms may be expected to adjust quickly. Scatterplots of this relationship at the sector and country level are shown in Figures 8 and 9, respectively. We find a strong relationship at the sectoral level. In general, manufacturing sectors that have greater volatility in output have weaker employment elasticities to growth. However, no clear pattern is evident at the country level (Figure 9). One possible reason for this is that countries end up adopting (maybe implicitly) a portfolio approach to their industry specialization, that balances high volatility sectors with others that are lower in volatility. 25 Figure 8 Relationship between sectoral output volatility and employment elasticity (contemporaneous) Figure 9 Relationship between national output volatility and employment elasticity (contemporaneous) HOW ARE EMPLOYMENT ELASTICITIES ASSOCIATED WITH OVERALL JOB CREATION? In this section, we ask if the employment elasticities we observed in the previous subsection are associated with aggregate job creation over a period. The point here, as was also discussed in section 2, is that having a robust relationship between growth and job creation is only useful if it is associated somehow with aggregate job creation. As a first step to exploring this issue, in Figure 10 we map countries in the dataset based on their average annual job creation during periods of growth and decline. We see significant variation across countries, although the most common situation is positive job creation in growth periods and job shedding in periods of decline. Here we have countries like Sri 26 Lanka, Libya, and to a lesser degree Jordan and Malaysia that experienced large jobs growth during growth periods but also relatively sharp job loss during declines. On the other hand, oil producing countries (UAE, Qatar) along with countries that have experienced strong structural transformation (China, Turkey, Honduras) experienced almost the same (large) shift both in periods of output growth and decline. Transition countries like Ukraine and Croatia show relatively strong downside responsiveness during decline but much weaker upside responsiveness during growth (this pattern holds for many transition countries). Figure 10 Relationship between size of employment movements during periods of growth and decline Figure 11 Relationship between elasticity and aggregate job creation 27 Figure 11 looks at long term job creation and its association with jobs elasticity of value added growth in the manufacturing sector. Here we measure overall job creation (growth %) over a period of 20 years7 and the average employment elasticity to value added growth over that period. Overall it shows a strong correlation between employment elasticity to growth and overall job creation. Again, we see the oil producers (UAE, Qatar, Kuwait, possibly Jordan) with very high overall job creation, despite responsiveness of employment to growth. Bangladesh, where growth and job creation has been rapid but where informality still dominates, looks similar. Finally, somewhat disparate set of countries, including Thailand, Ethiopia, Mexico, and Bolivia, show relatively strong responsiveness to growth along with robust job creation in the manufacturing sector Finally, we assess how the employment elasticity to growth varies by interacting the country income level and capital intensity of the subsector. Figure 12 shows that for low capital intensive activities, elasticity of jobs to growth is highest in low-income countries (as might be expected given their comparative advantage in labor costs. Similarly, the opposite is true for high income countries. However, as one moves to mid-level capital intensity, the relationship changes quickly with low and lower-middle income countries facing much lower employment elasticity, while is upper middle-income and high income countries experience a sharp jump in elasticity. Moving to highly capital intensive activities, however, we see elasticity jump almost all income groups, including the lowest income group. This dip in mid-tier levels of capital intensity and jump again at higher levels of capital intensity suggests the job creation relationship we are seeing may be linked to processes of structural transformation that emerge in the mid-levels of capital intensity. Yet at upper levels of capital intensity, activities may be of sufficiently higher value added or part of sufficiently large domestic clusters to make the employment impact of growth much stronger. These findings are supported in Figure 13, which shows the long-term employment impacts are actually highest (across just about all income groups) in the most capital intensive activities. 7 We take the latest 20 years available from each country, which results is significant differences in the periods covered across countries. 28 Figure 12 Elasticity of employment to growth by income and capital intensity level Figure 13 Long term job creation by income level and capital intensity 29 4. FIRM-LEVEL VIEW: FIRM TYPES, REGULATION, AND JOB CREATION In section 2, we explored the relation between changes in unemployment and GDP growth and determined that in high income countries there exists a statistically significant, negative association between these two variables. The higher the absolute value of the Okun’s coefficient, the more responsive is the labor market to changes in GDP. We also documented that countries with higher negative values of Okun’s coefficient, on average, also create more jobs during the period analyzed. In section 3, we tried to get a better perspective on these findings by considering the manufacturing sector, and found that the relationship between growth and job creation was strongest in the lowest and highest income countries, and in the least and most capital intensive sectors. Finally, in this section, as a complement to the previous analyses, we turn to the firm-level determinants of private sector job creation and carry out three separate, but related analyses. Firm entry and exit – Schumpeterian “creative destruction” – is critical for the continued dynamism of the modern economy. Though evidence has linked entrepreneurship and economic growth in developed countries, we have scarce evidence that such a relationship exists in developing countries. New firms enter the market and create new jobs, while other unprofitable firms exit the market contributing to job destruction (see e.g. Sutton (1997), Pakes and Ericson (1998), Geroski (1995)). Incumbent firms are in a continuous process of adaptation in response to the development of new products and processes, the growth and decline in markets and changes in competitive forces (Davis and Haltiwanger (1999)). Market conditions and institutional factors play a major role in shaping the magnitude of job flows and their characteristics (Davis et al. (1996)). In the first analysis in this section, using firm level data, we study how regulations affect employment growth across countries. We explore the links between the regulatory environment in which firms operate and job turnover by exploiting the observed industry-size variations through a difference-in- difference approach. Additionally, we explore firm characteristics that are associated with higher employment growth rates in the short run and the long run and identify those firms that create the largest impact on employment growth. We exploit cross-country differences to control for how macroeconomic conditions, and business and labor regulations affect job creation. Note that because we do not have access to cross-country panel data at the firm level, in this section we move away from the focus on responses to business cycles, although we do incorporate the Okun coefficient into the assessment. To carry out the analysis, we use surveys carried out by the world-wide Enterprise Survey project of the World Bank Group between 2006 and 2016. The firms surveyed are for the most part registered for tax purposes; they employ at least five employees and are located in urban areas. The manager/owner of the firm is interviewed in detail about basic firm characteristics (age, legal status, number of employees, etc.), as well as specific investment climate questions, for instance whether the firm has experienced power outages, what the delays have been when it requested a public service, how difficult it is to deal 30 with public officers, etc. The Enterprise Surveys use a stratified sampling methodology thus individual observations need to be properly weighted when making inferences about the population.8 THE REGULATORY ENVIRONMENT AND JOB TURNOVER With these data, we explore in detail the industry and size dimensions of the job flows, and relate them to institutional differences across countries. Two papers exploit job flows across industries within countries to investigate the role of employment protection: Haltiwanger, Scarpetta, and Schweiger (2008) and Micco and Pages (2006). Micco and Pages (2006) exploits industry level gross job flows data for manufacturing for 18 countries and uses a difference-in-difference. Haltiwanger, Scarpetta, and Schweiger (2008) use a harmonized firm-level database that covers all firms with at least one employee for both manufacturing and non-manufacturing sectors for 7 European countries, 5 Latin American countries, and 4 transition economies, and exploit country, industry and firm size variation in their analysis. The analysis hereafter follows closely the work of Haltiwanger et al. (2006) but we apply the methodology to 164 countries ranging from high-income countries to low-income countries. As in Haltiwanger et al. (2006), we find that firm size is by far the most important factor accounting for variation in the job flows across country, industry, and firm size classes. This suggests that exploiting data by firm size is important to provide greater within country variation in job flows for our empirical identification strategy, but also that distortions to job flows across countries may very well interact with the flow and firm size relationship. Specifically, we explore the links between the regulatory environment in which firms operate and job turnover by exploiting the observed industry-size variations through a difference-in-difference approach (see Rajan and Zingales (1998)). The test is constructed as follows: we identify an industry-size propensity for job reallocation from the Polish data. Poland is the country with the largest (negative) Okun coefficient as calculated in the analysis in Section 1, meaning that Poland has a labor market that adjusts easily to movements in business cycles.9 Under the assumption that regulations in the labor and goods markets in Poland are among the least restrictive in our sample, variation in job reallocation across industry-size cells in Poland should proxy for the technological and market driven differences in job reallocation in the absence of policy induced adjustment costs. Under the additional assumption that these technological and market driven differences in the demand for job reallocation carry over to other countries, we assess whether industry-size cells that have a greater propensity for job reallocation are disproportionally affected by regulations that raise adjustment costs. This would imply that, ceteris paribus, industry-size cells with more volatile idiosyncratic profit shocks and more frequent factor adjustments should be more strongly affected by regulations raising adjustment costs than those industry-size cells with less volatile idiosyncratic profit shocks and less frequent adjustment. The advantage of this approach compared to standard cross-country/cross-industry empirical studies is that we exploit within country differences between industry-size cells based on the interaction between 8 Under stratified random sampling unweighted estimates are biased unless sample sizes are proportional to the size of each stratum. With stratification, the probability of selection of each unit is, in general, not the same. Consequently, individual observations must be weighted by the inverse of their probability of selection (probability weights or pw in Stata.) 9 Haltiwanger et al. (2006) use the United States as the benchmark country for their sample. Since the enterprise Surveys mostly covers developing countries and does not include the United States we had to choose and alternative benchmark. As a robustness check we use Chile (best score in Economic Freedom Index in our sample) as a benchmark and results are consistent to those reported. These results are available from the author. 31 country and industry-size characteristics. Thus, we can also control for country and industry-size effects, thereby minimizing problems of omitted variable bias and other misspecifications. The core model specification used in our empirical analysis can be summarized as follows: 012345#6 = '7 + @×A 9 5#:. ,5# 85# + 6:. ,6 86 + '. (<32012345# ×>6 ) + /5#6 (8) where Dsi are industry x size si (si = 1,…., I x S) dummies, Dc are country c (c = 1,….,C) dummies, PolJFlowsi is Poland’s job flow variable in size class s and industry i, and / is the iid error term. Controlling for country effects sweeps out any country-specific variation, controlling for industry-size effects sweeps out the large common factors associated with industry and size, and the key interaction term between Poland’s flow in the industry-size class and the country regulation allows us to identify how the measured regulatory environment affects the variation across industry-size classes within countries. Poland’s job flow is used to quantify the propensity for the industry-size class to reallocate and, as discussed, reflects the fundamental driving forces underlying job reallocation across industry-size classes. In what follows, the measure of job flows used in the empirical analysis is the sum of job creation and H∈LM E∆FGHIJ job destruction rates (sum). Job creation rate is defined as C3D5#6% = and job destruction 7.O(FGHIJ EFGHI,JPQ) H∈LM E∆FGHIJ rate as RST5#6% = , where, i represents industry, s represents size class, c represents 7.O(FGHIJ EFGHI,JPQ) country, t represents time and E denotes employment. Capital letters S and C refer to a set of size classes or countries, respectively. The symbol ∆ denotes the first-difference operator, ∆U% = U% − U%-. .10 In addition to the core specification, we consider some closely related specifications. As a robustness check, we estimate an augmented model that also considers business sector regulations. We consider three measures of the regulatory environment for each country: one a composite index, one aimed at measuring labor regulations, and a third that measures business regulations. The source of these measures is the Economic Freedom of the World (EFW) measures created by the Frasier Institute. The index published in EFW, measures the degree to which the policies and institutions of countries are supportive of economic freedom. Studies have found that countries with institutions and policies more consistent with economic freedom have higher investment rates, more rapid economic growth, higher income levels, and a more rapid reduction in poverty rates. Each of these indexes is measure on a scale of 0 to 10, with 0 being the most restrictive environment. The average of the composite index is highest for high-income countries (7.18) and lowest for low-income countries (6.11) whereas the average for lower and upper middle-income countries falls in between these two groups. The monotonically increasing averages among income groups are also found for labor and business regulations. Previous research (see, e.g., Caballero et al. (2004), Heckman and Pages (2004)) suggests that regulations affect outcomes in the economy in the degree in which those regulations are enforced. Available indicators suggest a significant variation in the rules of law and the degree of enforcement of laws and regulations in our sample. Not only are some firms and jobs not registered in Latin America but 10 As in Haltiwanger et al. (2006), we take averages of pos and neg among firms, and then calculate sum. 32 also registered firms may also not fully comply with the existing rules and regulations where corruption is rampant—such as in some African countries in our sample. As an indication of the different degree of enforcement of laws and regulations, we consider the rule of law indicator from the World Bank’s World Governance Indicators. The indicator ranges between -2.5 and +2.5, with +2.5 being the best performer. The indicator shows high-income countries have the highest compliance with laws and regulations (average of 0.68), followed by upper middle-income countries (average of -0.37), then by lower middle- income countries (average of -0.67), and finally by low-income countries (average -0.83). To control for possibly differing degrees of enforcement of laws and regulations we adjust our regulatory variables as follows: Z[\] ^_ \W`Ea.O >6,WXY = ×>6 (9) O Table 6 presents the empirical results of the estimation of our core specification (8) and various variations. Recall that in equation (8) we specify a difference-in-difference analysis that identifies the impact of regulations via the interaction effect Poland’s job flows in the industry×size class with the country-specific regulation. The estimated coefficients on the interaction between Poland’s job flow and the different regulation measures (Table 6) is strongly significant overall, and in each of the income- groups when we allow the coefficient of the interaction to vary. Consistent with the working hypothesis, more volatile industries and firm size classes are impacted more in countries with more stringent regulations, including business and labor regulations. The results are robust to the different regulation variables adjusted for the degree of enforcement in each country. Table 6 Job Flows and the Role of Regulations (Difference in Difference Analysis) regulation = EF regulation = labor regulation = regulation = adj. EF regulation = adj. regulation = adj. VARIABLES index index business index index labor index business index (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) jflow_POL x regulation 0.064 0.037 0.039 0.029 0.043 0.029 [0.021]*** [0.011]*** [0.018]** [0.014]** [0.014]*** [0.015]* low income x jflow_POL x regulation -0.019 -0.016 -0.019 -0.042 -0.037 -0.044 [0.008]** [0.008]** [0.008]** [0.019]** [0.018]** [0.020]** lower-middle income x jflow_POL x regulation -0.014 -0.011 -0.013 -0.033 -0.030 -0.034 [0.006]** [0.006]* [0.007]* [0.016]** [0.017]* [0.017]* upperr-middle income x jflow_POL x regulation -0.007 -0.003 -0.006 -0.012 -0.005 -0.010 [0.006] [0.006] [0.007] [0.013] [0.013] [0.015] Constant -0.026 0.083 0.019 0.079 0.019 0.081 0.050 0.081 0.043 0.078 0.053 0.079 [0.051] [0.040]** [0.043] [0.040]** [0.046] [0.040]** [0.041] [0.040]** [0.041] [0.040]* [0.041] [0.040]** Observations 8,118 8,118 8,118 8,118 8,118 8,118 8,118 8,118 8,118 8,118 8,118 8,118 R-squared 0.229 0.229 0.229 0.228 0.228 0.228 0.228 0.228 0.229 0.228 0.228 0.228 Robust standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1. The depedent variable in all regressions is jflow sic . All regressions include sector x size effects and country effects. Determinants of Employment Growth at The Firm Level In this subsection, we explore specific firm characteristics that are associated with higher employment growth rates in the short run and the long run. We also exploit cross-country differences to discover how macroeconomic conditions, and business and labor regulations affect job creation. Following La Porta et al. (1999), Botero et al. (2004), and Fox and Oviedo (2013) we examine the link between regulation and job growth using a reduced-form model. We use robust OLS to estimate the 33 relationship between overall business and labor regulation and job growth, controlling for firm and country characteristics. We estimate the following equation: SbC_Td#% = +6 + +e + + % + 'fghS#%i + 'jTS# + 'k4RSdDℎgC# + mknℎSd# Where SbC_Td#% is firm i’s job (employment) growth; Size is the initial size of the firm, divided into six employment categories (5 or less, 6–10, 11–20, 21–50, 51–100 and over 100); Age is one of five age categories (less than 5, 5–9, 10–19, 20–39 and 40 and older); Ownership denotes whether the firm is private-domestic, private-foreign, public, or other. Other represents other firm characteristics, namely exporter status, access to credit, technology use, and human capital. We have two specifications, the first includes country-year and industry fixed effects. The second, includes, country, industry, and year effects. We include these effects to minimize the potential for omitted variable bias. The second specification allows us to control for country-year varying determinants such as GDP growth, the rolling Okun coefficient, and labor and business environment measured by EFW indicators described above. We cluster the errors at the country level. Finally, we estimated equation (1) using firm weights provided by the surveys. The dependent variable SbC_Td#% is measured either as annual average job growth over the last 3 years (short term) or as the average annual growth rate from startup to current size (long run) for firms that are at least 5 years old. In both cases, we have a truncation problem. The sample contains data on live firms only; therefore, changes in employment over the period considered for job growth exclude changes coming from firm exit. Obviously, the longer the period considered for job growth, the worse the truncation problem is likely to be. Thus, for the long-run job growth variable, the truncation problem may be severe. For example, if employment is more volatile in countries with a low EFW score than in countries with a high EFW score (there is more rapid expansion but also many more firm deaths), we would observe a higher expansion in countries with a low EFW score even though long-run job growth is actually higher in countries with a high EFW score because there are fewer deaths. Thus, these estimates should be regarded with care. Table 7 presents the OLS results for the two dependent variables. The results for the short-term growth confirm that smaller and younger firms grow faster, since all our size and age coefficients are negative and significant (the excluded categories are the smallest and youngest firms, respectively). The differences are greater between age classes than between size classes. Annual job growth seems to be closely related with exporter status and technology use, as measured by the use of e-mail. Also, firms with access to credit and that invest in human capital tend to have higher growth rates. These results— exporters and technology users that invest in human capital grow faster—are likely due in part to the fact that these types of firms are more productive. Regarding the country environment controls, we find that a more hassle-free labor environment is associated with higher employment growth, whereas countries with more hassle-free business regulations are related to lower employment growth. Even though institutions has a positive coefficient it is not statistically significant. We also control for macroeconomic variables including GDP growth, FDI, gross capital formation, and of particular interest, the Okun’s coefficient found in section 2. Results show that higher GDP growth, investment (i.e., gross fixed capital formation), and the level of development are all associated with higher levels of employment growth in the short run. FDI is found to have a negative and significant coefficient; however, it is not statistically significant (columns 2 and 3). A more responsive labor force, given by the 34 Okun coefficient, is associated with higher employment growth at the level of the firm; however, the effect is not statistically significant. Results from using long-term growth as the dependent variable differ and reveal further elements of employment dynamics. Micro-firms still grow faster (from a small base), but there are also large differences in growth between small, medium, and large firms. However, we find that state-owned enterprises have a negative effect on employment growth and the effect is statistically significant. In fact, all the firm characteristics included in the regression show positive and significant coefficients; this includes whether a firm provides training to their workers, suggesting that firms that invest in their workers have higher job growth in the longer term. Over the long-term, the macro variables, including the Okun coefficient, largely lose their statistical significance, except for openness which becomes statistically significant. In the long-run, it appears as a hassle-free labor environment deter employment growth. 35 Table 7 Determinants Employment Growth by Firms 1 2 3 4 VARIABLES short-run employment growth long-run employment growth exporter 0.013 0.008 0.006 0.009 [0.004]*** [0.004]** [0.004] [0.007] credit 0.009 0.020 0.005 0.016 [0.003]*** [0.002]*** [0.002]** [0.004]*** training 0.006 0.024 0.015 0.010 [0.004] [0.003]*** [0.003]*** [0.004]** email 0.006 0.015 0.011 0.008 [0.004] [0.002]*** [0.003]*** [0.004]** size 6-10 -0.040 -0.038 -0.021 -0.009 [0.006]*** [0.009]*** [0.003]*** [0.005]* size 11-20 -0.047 -0.043 -0.010 0.000 [0.006]*** [0.013]*** [0.004]*** [0.006] size 20-50 -0.051 -0.046 0.000 0.004 [0.006]*** [0.011]*** [0.003] [0.006] size 50-100 -0.059 -0.052 0.014 0.019 [0.006]*** [0.015]*** [0.005]*** [0.008]** size 100+ -0.061 -0.070 0.014 0.031 [0.007]*** [0.006]*** [0.005]*** [0.009]*** age 5-9 -0.042 -0.078 0.091 [0.011]*** [0.008]*** [0.004]*** age10-19 -0.058 -0.111 -0.050 -0.050 [0.011]*** [0.008]*** [0.001]*** [0.006]*** age 20 - 39 -0.065 -0.127 -0.077 -0.077 [0.011]*** [0.008]*** [0.001]*** [0.006]*** age 40+ -0.075 -0.140 -0.106 -0.111 [0.013]*** [0.009]*** [0.001]*** [0.007]*** majority foreign -0.004 -0.002 0.000 -0.000 [0.005] [0.005] [0.002] [0.009] majority state 0.003 -0.054 -0.021 -0.049 [0.012] [0.013]*** [0.004]*** [0.019]*** majority other 0.025 -0.010 -0.004 -0.018 [0.019] [0.011] [0.002] [0.009]** okun_chg_iunemp_rt -0.016 -0.004 [0.025] [0.028] polulation growth rate -0.079 -0.038 [0.024]*** [0.024] GDP growt rate 0.004 0.002 [0.001]*** [0.001] log GDPpc -0.022 -0.025 [0.066] [0.067] informality rate -0.005 -0.001 [0.002]** [0.001] domestic credit to private sector (% of GDP) 0.000 -0.000 [0.001] [0.000] gross fixed capital formation, private sector (% of GDP) 0.003 0.000 [0.001]** [0.001] foreign direct investment, net inflows (% of GDP) -0.004 -0.001 [0.001]*** [0.002] openess 0.000 0.001 [0.000] [0.000]** labor regulations 0.030 -0.017 [0.010]*** [0.010]* business regulations -0.044 -0.002 [0.013]*** [0.014] institutions 0.261 0.070 [0.210] [0.073] Constant 0.238 0.568 0.036 0.471 [0.031]*** [0.559] [0.015]** [0.575] Observations 92,450 27,890 79,222 23,383 R-squared 0.083 0.104 0.189 0.174 Robust standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1 36 Determinants of employment growth in high-impact firms In this final subsection, we focus the analysis on job creation from a small subset of firms – those referred to as ‘high-impact’ firms. While high-impact firms are seen to play an especially important role in the process of job creation over time compared with either the plants of large existing firms or the majority of very small startups that tend not to grow, to date there has been little research on the characteristics of high-impact firms, and even less on their micro and macro determinants (Acs, Parsons, and Tracy, 2008) With the framework that has been used throughout this study we analyze what determines the availability of high-impact firms across countries. We define a high-impact firm as a firm that has had impressive job creation in the three years leading up to the survey year. Specifically, a high-impact firm, is a firm that reaches an employment growth quantifier greater than 15. The employment growth quantifier (EGQ) is the product of the absolute and percent change in employment over a three-year period. The EGQ is used to mitigate the unfavorable impact of measuring employment change solely in either percent or absolute terms, since the former favors small companies and the latter large businesses. A firm with 10 employees in the initial year would have had to increase its number of workers to 23 in order to reach an EGQ higher than 15 whereas a firm with 100 initial employees would need to increase its labor force by 40 workers in order to reach a EGQ of 15. Annex Table 1 shows the number of firms by country-year in our sample that obtained an EGQ of 15, 20 and 50. Figure 14 shows the importance of high impact firms in an economy. Even though these firms represent a very small share of the number of firms in the economy, they create the majority of jobs. Firms with an EGQ of 15 or higher represent, on average, 5% of the firms in the sample whereas they create 44% of positive jobs in the economy. Clearly these shares vary by country and year, for example the share of firms in our sample which have an EGQ higher than 15, varies between 0 and 20% whereas the share of positive jobs created by firms with an EGQ higher than 15 ranges from 0 to 91%. This wide variation of the number of high performing firms across countries leads us to question why. What factors determine that a country produces a high number of high performing firms? Table 8 shows results of negative binomial regressions where the dependent variable is the number of firms with EGQ greater than 15, 20 and 50, and the independent variables are country level variables than aim at exploring the determinants of a proper environment for high-impact firms to flourish. Results show that a highly competitive environment, proxied by the number of firms, economic growth, and a responsive labor market proxied by the Okun coefficient are the three most robust factors for an environment that produces high impact firms. Investment by the private sector, openness, flexible labor and business regulations, and rule of law that protects private rights are all positively related with higher number of high impact firms; however, their effect is no statistically significant. Counterintuitively, foreign direct investment and access to credit have a negative effect; however, it is not statistically significant. 37 Figure 14 High Impact Firms - Average Share of Firms and Average Share of Positive Jobs 44.4 40.9 40 30.3 30 20 10 5.4 4.3 2.0 0 share of firms HIMPF 15 share of postive jobs HIMPF 15 share of firms HIMPF 20 share of postive jobs HIMPF 20 share of firms HIMPF 50 share of postive jobs HIMPF 50 Table 8 Determinants of High Impact Firms (1) (2) (3) VARIABLES himpcf_15 himpcf_20 himpcf_50 log number of firms 1.007 1.058 1.079 [0.061]*** [0.067]*** [0.084]*** okun (difference) -0.645 -0.580 -1.325 [0.172]*** [0.197]*** [0.284]*** polulation growth rate -0.031 -0.053 -0.047 [0.023] [0.027]** [0.039] GDP growt rate 0.025 0.024 0.021 [0.008]*** [0.010]** [0.014] log GDPpc 0.139 0.051 0.157 [0.106] [0.119] [0.181] domestic credit to private sector (% of GDP) -0.002 -0.002 -0.006 [0.003] [0.004] [0.005] gross fixed capital formation, private sector (% of GDP) 0.022 0.022 0.015 [0.013]* [0.016] [0.018] foreign direct investment, net inflows (% of GDP) -0.039 -0.055 -0.050 [0.026] [0.033]* [0.034] openess 0.000 0.001 0.002 [0.003] [0.004] [0.005] labor regulations 0.010 0.006 0.009 [0.048] [0.053] [0.085] business regulations 0.186 0.152 0.190 [0.082]** [0.094] [0.133] institutions 0.270 0.156 0.200 [0.157]* [0.178] [0.242] Observations 90 90 90 Robust standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1 38 5. CONCLUSIONS AND FUTURE RESEARCH This paper is intended as the first in a series of assessments that aim to understand better the patterns and determinants of private sector job creation. In this first paper, we took an exploratory look at the relationship between growth and jobs by analyzing patterns of job creation and destruction in business cycles across a large number of global economies. The objective of the exercise was modest – to assess the patterns of the growth-jobs relationship by countries, income groups, economic structures, sectors, and firm types. Three separate, but related analyses were carried out, looking at the macro, sectoral, and firm levels. From the descriptive analysis, we found that while growth is, of course, fundamental for job creation, very different labor market responses to growth emerged across countries and income groups. Taking only the macro view, we saw that low-income countries, particularly when dominated by agricultural and other forms of self-employment, appear to have job creation patterns that are unresponsive to growth. However, the industry-level view, which largely controls for self-employment, suggests that employment in low-income countries does in fact respond elastically to growth, at least in the formal sector. These findings highlight the important role of structural transformation in shaping the growth- jobs relationship. More broadly, the significant differences in the patterns of job creation across income groups and economic structures calls for matters for considering how and when we should expect jobs growth at different stages of the economic cycle. An important focus of this paper was to test whether labor market responsiveness to growth and decline (as measured by the Okun coefficient or by standard elasticities) is associated with overall job creation. Here we found that economies where formal sector job creation was more responsive to growth cycles (i.e. they create jobs in line with growth and shed jobs in line with decline) generated more jobs overall than economies with less responsive patterns of job creation. This was also found to be true at the firm level – firms which create and shed jobs more fluidly create more jobs overall. At the same time, many countries (and presumably, many firms) expanded employment significantly in aggregate despite a relatively inelastic response to business cycles; similarly, some countries exhibited weak overall job creation despite a highly responsive labor market. This appears to be driven mainly by structural factors – less responsive, high job-creating countries consist mainly of: i) high income, extractive (mainly oil producing) economies; and ii) high-growth, early-stage transformation (high informality) economies. By contrast highly responsive, low job-creating countries appear to be mainly transition economies in Central and Eastern Europe, where the long-run analysis picks up major declines in employment from the 1990s. Another important finding in the paper relates to job creation in technology and capital intensive environments. The results from both the macro analysis (grouping of ‘high technology’ countries) and the sectoral analysis indicate that, far from being a simple substitute from jobs, economies and sectors that are capital intensive tend to be have a responsive growth-jobs relationship. In fact, the sectoral analysis in manufacturing found highly capital intensive sectors were associated with higher employment elasticity to growth than medium or low capital intensive sectors across most income levels; and for all income levels, highly capital intensive sectors generated faster overall employment 39 growth than less capital intensive sectors. This suggests significant complementarity between capital and labor, a finding which certainly should be probed more deeply. Finally, while the focus of the paper was to take a broad look at economic growth and job creation relationships across a wide range of countries, we also carried out a broad assessment on the determinants of job creation at the firm level. Any analysis on the determinants of a fundamental economic outcome like job creation faces huge challenges of endogeneity. Nevertheless, we were able to identify some factors that are at least correlated with job creation outcomes. First, in terms of firm types, the analysis supports much of the recent research showing that small and (especially) young firms are associated with greater job creation. More broadly, by far the most important determinants of job creation at the firm level are either firm-specific characteristics (including demographics, ownership, and firm behaviors like use of technology and training) or macro-economic and structural factors (including growth levels, investment, and labor market responsiveness). Beyond these (and another, arguably structural, factor – institutional quality), access to formal bank financing, the business regulatory environment, and some limited aspects of labor regulations, appear to be relevant. Indeed, while business regulatory issues do not appear to be among the most fundamental factors shaping job creation outcomes, they may be important at the margin – the evidence presented in this paper identifies business regulation as particularly significant in industry sectors and firm size classes that face greater volatility in terms of profits and factor adjustments. Together, these findings point to the importance of market rigidities that may constrain the ability of firms to respond to growth opportunities. Finally, the analysis of high impact firms suggests the importance of the competition environment in the emergence of highly productive and job-creating firms. While the analysis on determinants of firm-level job creation presented here is only at a very high level, the above findings may help to frame a potential policy agenda for job creation, including: macro- economic fundamentals, responsive labor markets, access to finance, competition, and a facilitative business regulatory environment. These are not surprising, but nevertheless frame a set of issues that could be explored in future papers in this series. Indeed, bearing in mind the exploratory nature of the paper, the results are promising and suggest fertile ground for further analysis. Future papers in the series may consider two main avenues to deepen this initial analysis on economy wide job creation: first, exploring how sectoral and enterprise structures shape patterns of job creation; second, assessing in more detail the determinants of factor intensity and job creation at the firm level. On the former, an immediate follow-on to the findings in this paper may be to probe job creation patterns in a set of country case studies. Such an approach would allow for more detailed quantitative assessment, including linking job creation outcomes with specific policy actions. These case studies would focus on countries that appear as outliers to the patterns discussed in this paper, or where there are particularly interesting episodes of growth and decline (e.g. surges and slumps). This may allow us to identify more specifically the dynamics job creation / job shedding responses, and the timing of those responses, in key stages of the business cycle (entering or coming out of a slump) impact longer-run outcomes. Another potentially valuable follow-on from the findings in this paper would involve assessing in more detail how sectoral specialization impacts job creation patterns. Such an assessment might follow along the lines of Loayza and Raddatz (2006), who found that the sectoral composition of growth (and, 40 specifically, relative labor intensity) had important implications for the impact of growth on poverty. Another potentially valuable follow-up. Further down the line, future research following the results of this paper may involve the challenging task of incorporating earnings into the analysis to allow for a more nuanced understanding of economies characterized by high levels of self-employment and informality, as well as identifying distortions that may lead to a systematic underutilization of labor in some economies. Finally, much work remains to understand firm-level determinants to job creation. Within this, a high priority in this series of papers on private sector job creation is to deepen our knowledge on the factors which determine whether and how a firm chooses to employ labor, particularly with respect to capital, when facing a known (or, conversely an uncertain) future pattern of growth. 41 ANNEX A: OKUN’S LAW The basic bivariate model specification that was used by Okun (1962), inter alia, by Weber (1995) may be represented by the following set of equations: 6 ∗ (#,% ≡ (#,% − (#,% (1) 6 ∗ "#,% ≡ "#,% − "#,% (2) 6 6 "#,% = '(#,% + /#,% (3) 6 where (#,% is the logarithm of cyclical output, the output gap; (#,% is the logarithm of observed or actual ∗ 6 output; (#,% is the logarithm of potential or natural output; "#,% is the cyclical unemployment rate, the ∗ unemployment gap; "#,% is the observed or actual unemployment rate; "#,% is the potential or natural unemployment rate; and /#,% is a stochastic error term. The subscript i denotes country i in year t. Okun’s coefficient is measured by the estimated value of ' , the impact coefficient, such that ' ≤ 0. Obviously, ∗ ∗ to estimate this coefficient we need time series on the unobserved components "#,% and (#,% . The trend values are found through a linear regression using two lags of the corresponding variable as Hamilton (2017) recommends not using Hodrick-Prescott (HP) filter as it produces series with spurious dynamic relations that have no basis in the underlying data-generating process and suggest using variables lags instead as a cleaner detrending process. We also use a moving average of the form (% ∗ = ((%E. + (% + (%-. )/3 to find the natural output and employment. Since determining the natural rate of growth in this framework is non-trivial for our results we decided to use the difference specification to make sure that the detrending of the series was not the driving force in our results. However, we make use of the gap specification to check the robustness of our results. Table 1 shows results using Okun’s coefficients estimated by the gap specifications using the two detrending methodologies. In this Annex, we test some alternative specifications of the Okun model, mainly the idea that the relation between unemployment and output growth does not necessarily need to be linear. Annex Figure 1 shows the difference between the fitted values estimated by a linear and by a non-linear regression model for the cases of the United States, Ecuador, Tanzania, and Vietnam. We see that the non-linear model displays some interesting patterns that are not accounted for in the linear model. For example, the data for the United States displays a convex shape whereas Ecuador, Tanzania, and Vietnam have a concave form. In the case of the United States, the rate at which unemployment decreases, slows down as GDP grows at faster rates. On the other hand, the developing countries’ concave form, suggest that unemployment actually increases with low level of positive growth and only becomes negative once a faster rate of growth in achieved. The convex form in the US is probably related to the country reaching its natural rate of unemployment, as once it is reached, more output growth will not lead to less unemployment. Even though these patterns are extremely interesting and there is more to be explored, the issue is that we now have two coefficients, rather than one, and how to we explore the Okun coefficient as an employment output when there is more than one part to control for. However, it is important to highlight that both linear and non-linear Okun coefficients are highly correlated. 42 Annex 1 - Figure 1 Non-Linear Okun 43 Annex 2 - Figure 2 Distribution of Okun’s coefficient by income group 2.5 6 2 4 1.5 Density Density 1 2 .5 0 0 -1 -.8 -.6 -.4 -.2 0 -.3 -.2 -.1 0 .1 okun_b okun_b 2.5 10 2 8 1.5 Density Density 6 1 4 .5 2 0 0 -.5 0 .5 1 -.2 -.1 0 .1 .2 okun_b okun_b 44 Annex 3 - Table 1 High Impact Firms by Country share of firms share of positive jobs country year all firms himpcf_15 himpcf_20 himpcf_50 jobs himpcf_15 himpcf_20 himpcf_50 Angola 2006 834 3.8 3.1 1.0 5,006 13.8 11.2 3.4 Angola 2010 618 11.9 8.9 3.9 5,418 47.7 42.2 32.3 Albania 2007 2,126 9.0 6.4 0.9 12,672 40.7 27.5 10.4 Albania 2013 736 2.6 2.3 0.5 3,010 38.5 35.2 23.2 Armenia 2009 1,098 14.1 9.1 3.5 19,905 58.9 52.6 41.6 Armenia 2013 1,439 4.3 4.1 1.6 7,695 58.1 57.3 28.7 Burundi 2014 661 5.2 3.0 0.4 6,543 25.8 15.9 11.8 Benin 2009 1,577 8.4 6.1 3.1 4,575 44.4 38.8 24.2 Burkina Faso 2009 1,655 3.2 2.7 1.2 10,116 37.7 36.3 28.5 Bangladesh 2007 26,829 5.8 4.3 2.7 399,710 69.9 62.6 51.6 Bangladesh 2013 7,319 11.0 10.2 7.1 344,918 86.5 85.6 79.5 Belize 2010 759 0.3 0.0 0.0 1,475 8.8 0.0 0.0 Bolivia 2006 6,086 10.3 9.2 2.9 41,750 44.9 42.3 18.6 Bolivia 2010 4,063 4.5 4.4 1.0 27,514 37.4 36.9 19.8 Brazil 2009 292,663 12.2 11.0 6.5 4,655,269 78.8 76.6 69.5 Botswana 2006 1,166 6.6 5.6 1.0 10,565 16.8 15.7 6.4 Botswana 2010 675 9.2 7.9 3.7 9,275 79.1 77.0 66.5 Côte d'Ivoire 2009 8,998 4.0 2.6 1.2 39,355 21.9 17.3 13.6 Cameroon 2009 1,329 3.4 3.0 2.2 9,245 62.9 59.2 49.9 Congo, Rep. 2009 1,423 6.1 1.7 1.0 8,775 31.1 23.9 17.3 Ecuador 2006 6,185 7.6 6.6 3.7 60,147 54.3 51.5 40.9 Ecuador 2010 7,830 2.8 2.5 1.0 64,564 36.7 34.5 22.8 Egypt, Arab Rep. 2013 114,226 2.6 2.2 1.8 818,282 42.0 40.4 37.9 Georgia 2008 3,122 18.9 14.2 9.6 108,718 91.0 89.3 81.8 Georgia 2013 7,497 11.3 10.7 7.2 117,155 68.2 67.9 64.6 Ghana 2007 6,386 6.2 4.8 2.1 50,816 52.3 47.9 39.0 Ghana 2013 1,371 4.9 3.3 1.3 7,971 40.6 31.7 18.1 Guatemala 2006 14,234 7.4 3.7 1.4 101,025 53.5 40.8 28.6 Guatemala 2010 9,629 2.7 2.5 1.6 105,572 65.8 65.5 58.4 Guyana 2010 321 7.1 4.9 2.5 3,324 45.1 40.2 26.8 Honduras 2006 7,653 13.2 12.7 8.1 145,346 80.0 78.9 71.5 Honduras 2010 5,122 3.4 3.4 0.5 27,545 66.7 66.5 26.4 Croatia 2013 11,944 3.8 3.0 2.3 54,601 44.4 42.9 35.5 Jordan 2013 8,396 3.3 2.7 0.6 55,407 47.5 45.3 36.3 Kyrgyz Republic 2009 1,002 9.9 9.0 1.8 11,652 41.6 40.2 18.7 Kyrgyz Republic 2013 939 3.8 1.2 0.8 8,131 50.4 38.5 34.4 Lebanon 2013 6,348 6.2 5.2 2.4 32,671 51.2 47.3 29.9 Madagascar 2009 1,891 5.5 4.3 2.8 23,571 44.4 42.9 39.7 Madagascar 2013 2,434 3.0 3.0 2.2 29,154 55.8 55.8 52.7 Mexico 2006 76,686 4.3 3.5 1.4 515,381 53.3 50.7 36.7 Mexico 2010 68,465 4.4 2.7 1.3 652,430 63.7 62.1 56.4 Macedonia, FYR 2009 4,935 12.4 9.6 6.7 43,169 42.3 40.4 32.7 Macedonia, FYR 2013 3,056 3.0 2.5 1.2 11,789 44.2 40.1 25.8 Mali 2010 970 1.1 0.8 0.2 1,729 18.6 17.2 8.4 Montenegro 2013 1,965 0.0 0.0 0.0 2,422 0.0 0.0 0.0 Mongolia 2009 2,567 18.5 14.6 5.5 30,445 66.7 61.7 40.0 Mongolia 2013 4,769 3.8 0.4 0.1 28,673 36.7 10.5 6.8 Mauritius 2009 4,238 8.6 6.8 3.5 60,949 79.3 77.8 66.1 Malawi 2009 1,778 11.7 7.4 3.9 26,585 70.6 67.4 58.9 Malawi 2014 2,924 3.5 3.3 1.3 71,763 86.0 85.7 81.5 Namibia 2014 7,119 3.1 2.2 0.7 18,026 23.7 22.0 15.8 Niger 2009 921 5.1 2.9 0.5 4,010 30.4 23.5 15.0 Nicaragua 2010 1,878 3.8 2.3 0.7 11,505 49.1 46.3 27.0 Nepal 2009 11,930 2.7 2.5 0.2 36,213 19.4 18.2 6.3 Nepal 2013 9,877 2.8 2.8 0.1 22,300 34.7 33.6 5.9 Pakistan 2007 25,114 2.2 1.8 0.5 104,299 52.3 49.7 36.9 Pakistan 2013 12,549 3.7 3.4 1.9 196,698 64.1 63.0 53.5 Panama 2006 3,437 4.1 2.7 1.6 18,463 50.2 40.6 31.2 Panama 2010 5,359 7.7 5.8 4.5 37,763 51.8 46.4 38.5 Peru 2006 6,076 18.9 14.5 12.3 159,257 80.0 76.0 71.7 Peru 2010 13,712 9.5 8.9 1.7 160,409 71.8 70.0 41.4 Philippines 2009 33,366 6.3 5.8 2.5 379,664 81.6 80.0 65.5 Paraguay 2010 2,144 10.7 10.2 5.3 29,250 68.2 67.6 56.3 Russian Federation 2009 84,647 14.9 14.1 8.5 3,940,826 88.8 88.0 83.8 Russian Federation 2012 83,316 6.5 5.3 2.6 663,197 48.0 45.9 34.3 Rwanda 2006 598 8.0 6.4 2.4 5,776 27.1 25.8 13.8 Senegal 2007 1,714 3.2 2.2 1.0 8,616 24.2 22.1 10.6 Senegal 2014 2,606 1.6 1.4 0.3 11,287 18.6 17.5 11.7 Sierra Leone 2009 870 4.4 2.5 0.6 2,704 38.7 35.0 22.4 El Salvador 2006 10,287 9.0 6.6 2.0 78,474 59.7 53.1 37.0 El Salvador 2010 4,809 3.1 1.8 1.3 23,478 34.6 23.5 16.7 Serbia 2013 14,762 0.7 0.7 0.3 51,145 19.3 17.0 10.1 Suriname 2010 274 1.7 1.1 1.1 1,022 43.4 39.1 39.1 Chad 2009 647 7.0 3.9 0.9 3,943 25.8 16.2 7.2 Togo 2009 1,382 2.0 2.0 0.0 4,710 6.8 6.8 0.0 Tajikistan 2013 1,240 6.3 4.1 2.7 11,581 42.4 36.7 32.5 Trinidad and Tobago 2010 5,852 1.1 0.5 0.4 15,711 12.6 4.7 3.9 Turkey 2008 43,680 15.4 12.5 6.9 1,220,251 87.0 85.5 79.0 Turkey 2013 103,852 7.5 7.1 3.2 597,295 56.6 54.7 30.2 Tanzania 2006 7,293 6.9 5.0 1.8 59,191 49.1 45.1 30.6 Uganda 2006 3,969 6.5 4.4 1.1 24,237 39.8 34.4 16.0 Uganda 2013 6,752 2.5 2.5 0.3 17,119 35.8 35.4 16.0 Ukraine 2008 48,119 7.8 6.3 2.4 445,132 48.0 44.3 33.4 Uruguay 2006 5,492 7.5 6.1 2.9 37,881 42.0 38.0 23.7 Uruguay 2010 3,385 2.2 1.9 0.3 17,246 39.8 36.9 18.8 Venezuela, RB 2006 18,552 7.3 5.8 2.1 162,155 37.6 35.3 24.4 Venezuela, RB 2010 10,910 3.5 3.5 2.1 56,074 75.3 75.3 69.0 Yemen, Rep. 2010 5,934 3.2 2.8 0.5 17,279 44.0 41.7 23.8 Yemen, Rep. 2013 5,273 1.7 1.3 0.6 10,849 52.5 48.2 36.1 South Africa 2007 120,831 7.0 4.4 1.7 1,288,628 31.2 26.5 13.7 45 BIBLIOGRAPHY Acs, Zoltan J., William Parsons, and Spencer Tracy. 2008. “High-Impact Firms: Gazelles Revisited.” unpublished manuscript prepared for the United States Small Business Administration. Ahmed, Masood, Dominique Guillaume, and Davide Furceri. 2012. “Youth Unemployment in the MENA Region: Determinants and Challenges.” in Addressing the 100 Million Youth Challenge: Perspective on Youth Employment in the Arab World in 2012. Geneva: World Economic Forum. Ball, Laurence, Daniel Leigh, and Prakash Loungani. 2016. “Okun’s Law: Fit at 50?” Revised version of NBER Working Paper No. 18668. Botero, J., S. Djankov, R. Porta and F. C. Lopez-De-Silanes. 2004. “The Regulation of Labor.” The Quarterly Journal of Economics. 119 (4): 1339–82. Caballero, Ricardo J., Kevin N. Cowan, Eduardo M.R.A. Engel, and Alejandro Micco. 2004. “Effective Labor Regulation and Microeconomic Flexibility.” Discussion Paper 1480. Cowles Foundation. Davis, Steven J. and John C. Haltiwanger. 1999. “On the Driving Forces Behind Cyclical Movements in Employment and Job Reallocation.” American Economic Review, 89(5):1234-1258. Davis, Steven J., John C. Haltiwanger, and Scott Schuh. 1996. Job Creation and Destruction. The MIT Press: Cambridge, Massachusetts. Decker, Ryan, John Haltiwanger, Ron Jarmin, and Javier Miranda. 2014. "The Role of Entrepreneurship in US Job Creation and Economic Dynamism." Journal of Economic Perspectives. 28(3): 3-24. Fort, Teresa, John Haltiwanger, Ron S. Jarmin, and Javier Miranda. 2013. “How Firms Respond to Business Cycles: The Role of the Firm Age and Firm Size.” IMF Economic Review. 61: 520-559 Fox, Louise, and A. M. Oviedo. 2008. “Are Skills Rewarded in Sub-Saharan Africa? Determinants of Wages and Productivity in the Manufacturing Sector.” World Bank Policy Research Working Paper WPS4688. Washington, D.C.: World Bank. Freund, Caroline and Bob Rijkers. 2014. “Episodes of Unemployment Reduction in Rich, Middle-Income, and Transition Economies.” Policy Research Working Paper No. 6891. Washington, D.C.: World Bank. Geroski, Paul A. 1995. “What do We Know about Entry?” International Journal of Industrial Organization. 13:421-440. Gertler, Mark and Simon Gilchrist. 1994. “Monetary Policy, Business Cycles, and the Behavior of Small Manufacturing Firms.” The Quarterly Journal of Economics. 109(2): 309-340. González, Alvaro S., Leonardo Iacovone, and Hari Subhash. 2013 “Russian Volatility: Obstacle to Firm Survival and Diversification.” World Bank Policy Research Working Paper No. S6605. Washington, D.C.: World Bank. Gordon, Robert J. 1984. ‘‘Unemployment and Potential Output in the 1980’s.’’ Brookings Papers on Economic Activity. 537–564. 46 Haltiwanger, J., S. Scarpetta and H. Schweiger. 2006. “Assessing Job Flows across Countries: The Role of Industry, Firm Size, and Regulations.” World Bank Policy Research Working Paper No. 4070. Washington, D.C.: World Bank. Hanusch, Marek. 2012. “Jobless Growth? Okun’s Law in East Asia.” World Bank Policy Research Working Paper No. 6156. Washington, D.C.: World Bank. Heckman, James J. and Carmen Pages (editors). 2004. Law and Employment: Lessons from Latin America and the Carribean. NBER Conference Series Report. University of Chicago Press, Chicago and London. Hendry, David F., and Grayham E. Mizon. 1978. ‘‘Serial Correlation as a Convenient Simplification, Not a Nuisance: A Comment on a Study of the Demand for Money by the Bank of England.’’ Economic Journal. 88: 549–563. Kaufman, Roger T. 1988. ‘‘An International Comparison of Okun’s Law.’’ Journal of Comparative Economics. 12: 182–203. Knoester, Anthonie. 1986. ‘‘Okun’s Law Revisited,” Weltwirtschaftliches Archiv Review of World Economics. 122(4):657–666. Knoteck, Edward S., II. 2007. “How Useful is Okun’s Law?” Federal Reserve Bank of Kansas City. Loayza, Norman V. and Claudio Raddatz. 2006. "The composition of growth matters for poverty alleviation." World Bank Policy Research Working Paper WPS4077. Washington, D.C.: World Bank La Porta, R., F. Lopez-De-Silanes, A. Shleifer and R. Vishny. 1999. “The Quality of Government.” Journal of Law, Economics and Organization. 15 (1): 222–79. Micco, Alejandro and Carmen Pages. 2006. “The Economic Effects of Employment Protection: Evidence from International Industry-Level Data.” Discussion Paper 2433. IZA. Mizon, Grayham E. 1995. “A Simple Message for Autocorrelation Correctors: Don’t." Journal of Econometrics. 69: 267–288. Moosa, Imad A. 1997. “A Cross-Country Comparison of Okun’s Coefficient.” Journal of Comparative Economics. 24(3): 335–56. Okun, Arthur M. (1962). “Potential GNP: Its Measurement and Significance,” American Statistical Association, Proceedings of the Business and Economics Statistics Section, pp. 98-104 Pakes, Ariel, and Richard Ericson. 1998. “Empirical Implications of Alternative Models of Firm Dynamics.” Journal of Economic Theory. 79(1):1-45. Prachowny, Martin F. J. 1993 ‘‘Okun’s Law: Theoretical Foundations and Revised Estimates.’’ Review of Economics and Statistics. 75: 331–336. Rajan, Raghuram G. and Luigi Zingales. 1998. “Financial Dependence and Growth.” American Economic Review. 88(3):559-586. Smith, Gary. 1975. ‘‘Okun’s Law Revisited.’’ The Quarterly Review of Economics and Business. 37–54. Winter 1975. 47 Sutton, John. 1997. “Gibrat's Legacy.” Journal of Economic Literature. 35(1):40-59. Weber, Christian E. 1995. ‘‘Cyclical Output, Cyclical Unemployment, and Okun’s Coefficient: A New Approach.’’ Journal of Applied Econometrics. 10: 433–445. World Bank. 2013. World Development Report 2013: Jobs. Washington DC: World Bank. 48